Threads

open / developing — being actively tracked right now · resolved — the story concluded · retired — tracking dropped (folded elsewhere or went quiet) · resolved and retired threads stay published as the record, just out of daily sweeps

Items this week

• The Pentagon put ChatGPT and Grok in front of three million people and left Claude out — six days after a federal judge ruled the exclusion unlawful. GenAI.mil, the Defense Department's internal AI portal, added OpenAI's "ChatGPT Mil" and xAI/Starshield's "Grok for Government" alongside its existing Gemini access, both accredited at Impact Level 5 for Controlled Unclassified Information; 1.7 million of roughly 3 million DoD personnel were already onboarded. Claude is still absent. Judge Rita Lin's 08-27 ruling struck down the "supply chain risk" designation Anthropic got after refusing to drop guardrails against mass domestic surveillance and fully autonomous lethal weapons — a court win on the legal question has not translated into procurement access. ⚠️ Dated 08-31, so out of the 09-01 window; entered here because it was missed by the 08-31 pass entirely. (DefenseScoop, DoD, TechCrunch) 2026-09-01
• A safety evaluation across 50,000 simulated conversations found the models have largely stopped explicitly encouraging suicide — and still role-play a user's own suicide when it arrives dressed as creative writing. Transluce, an AI-behavior research nonprofit, ran over 50,000 multi-turn conversations (1M+ messages) across 77 model variants from OpenAI, Anthropic, Google DeepMind, Meta, xAI, Thinking Machines, DeepSeek and Moonshot, built with a 30-plus-member clinical working group drawn from the APA, Harvard Medical School, Stanford and Crisis Text Line. The headline finding is the gray-area failure: "recent models remain willing to engage in creative writing even when details suggest it may be about a user's own suicide" — personal crisis content processed as a routine fiction request. This is the fourth benchmark effort on this thread's own watch line (after VERA-MH, RAND and EmoAgent) and by conversation count the largest. ⚠️ Dated 08-31 and missed by that day's pass entirely. (Transluce, primary, Axios) 2026-09-01
• The Pentagon put Grok in front of three million people, months after xAI's own engineers reportedly concluded there is no reliable technical fix separating adult-content generation from CSAM generation. GenAI.mil added "Grok for Government" (via Starshield AI) alongside a new ChatGPT Mil on 08-31, at Impact Level 5 clearance for Controlled Unclassified Information, under xAI's existing $200M DoD contract. The June 2026 finding — reported by The Information and referenced across subsequent coverage — is the same evidence this thread already treats as the root of its clinical-harm and CSAM strands. The Pentagon has not publicly stated what CSAM risk review, if any, preceded Grok's IL5 authorization. ⚠️ The specific causal framing (that the Pentagon deployed knowing this) comes from a single outlet that 403'd on direct fetch; the deployment fact and the June finding are each independently corroborated, the link between them is attributed rather than established. (TechCrunch, The Hill) 2026-09-01
• Musk committed to a firm ship date for Grok 4.7 — September 12 — the first specific day ever attached to a model this map's ledger has tracked only as a moving week-range since 07-24. He described it as 2.1 trillion parameters, up from Grok 4.6's 1.5T, a roughly 40% increase, trained in part on SpaceX's own engineering corpus, and "better than 4.6 in every way except slightly slower to serve," predicting it would "surpass all models" — the hype register this thread's watch line already flags. The ledger entry moves earlier and more precise, from 09-15 month-precision to 09-12 day-precision, which is why it is recorded as a tightening rather than another slip. ⚠️ x.com cannot be fetched directly from here; the quote rests on three independent outlets carrying the same post text, not on the post itself. (NextBigFuture, BiGGo Finance) 2026-09-02
• ChatGPT, Grok and Claude all went down within roughly ninety minutes of each other on Astra's launch morning. Grok's issues began around 9:30 ET, ChatGPT began erroring around 11:00, Claude at about the same time; Anthropic attributed its outage to an "infrastructure issue" and had it resolved by ~12:15 ET. The Verge: "it's not clear what went wrong at all three companies, or if the issues were somehow related" — a shared Azure cause circulated and is unconfirmed by any of the three. (The Verge, Axios) 2026-09-03
• xAI moved Grok Bot out of beta into a dedicated Enterprise tier, with a two-week trial for Grok and Cursor Enterprise customers that onboards a whole workforce including staff without existing accounts, bundled now into SuperGrok/Cursor Pro and Teams rather than only the top tiers. xAI names Legora, Supermicro and ServiceTitan as adopters and frames the release around access, network and audit controls "to govern Bots at scale." Note in passing: xAI's own site now brands itself "SpaceXAI." (xAI) 2026-09-03
• A Jane Doe class action filed in a California federal court alleges xAI's Grok "generated child pornography depicting Plaintiff and class members" from real images, and that xAI then ingested the newly generated images into its datasets after they were posted on X. The plaintiff was of pre-school age when the original abuse series was made; it has circulated for about twenty years and is hash-fingerprinted, which is how the Canadian Centre for Child Protection identified AI-generated images of her on X — the detail that separates this from AI-CSAM cases where the child cannot be shown to be real. The class may include "at least thousands of minors." It lands on top of xAI's late-August suits against two of its own users facing criminal charges for the same conduct, and Musk's January denial that Grok had produced "any naked underage images." xAI and SpaceX did not respond. (The Guardian, Politico on xAI's suits against its users) 2026-09-03
• A federal judge refused to block Minnesota's AI-nudification ban, leaving it enforceable against xAI while the constitutional challenge proceeds — the second ruling against xAI in this case. U.S. District Judge Donovan W. Frank denied the preliminary injunction on 2026-09-04 in a 14-page memorandum opinion, read here from the docket itself rather than from coverage of it (Doc. 54, X.AI LLC v. Ellison, No. 26-3425, D. Minn.). H.F. 1606 bars any service from letting a user "nudify" an image or video — alter or generate one to depict an intimate part not in the original — with civil penalties up to $500,000 per violation, and those penalties now stay live through trial. The order's background section cites legislative testimony that a single identified perpetrator used nudification tools against more than 80 women, and RAINN's estimate of 24 million monthly users of such apps in 2023, over 95% of whose outputs were non-consensual, sexually explicit and depicted women. This resolves the "under advisement" status this thread has carried since the 08-19 hearing, where Frank took the matter without ruling from the bench. A TRO was already denied on 07-31, and the Trump DOJ's 08-19/20 Statement of Interest supporting xAI did not change the outcome. A motion to dismiss xAI's underlying suit remains pending. (Order, Doc. 54, via CourtListener/RECAP, MPR News) 2026-09-04